Closed tbates closed 4 years ago
Test example, just adding noise to Twin 2 moderator ...
tmp = twinData
tmp$age1 = twinData$age
tmp$age2 = tmp$age1 +rnorm(n=length(tmp$age1))
selDVs = "bmi"; selDefs = "age"
mzData = subset(tmp, zygosity == "MZFF"); dzData = subset(tmp, zygosity == "DZFF")
m1 = umxGxE(selDVs= "bmi", selDefs= "age", sep= "", dzData= dzData, mzData= mzData, tryHard= "yes")
name | Estimate | Std.Error |
---|---|---|
a_r1c1 | 0.700 | 0.041 |
c_r1c1 | 0.000 | 0.342 |
e_r1c1 | 0.293 | 0.030 |
am_r1c1 | 0.002 | 0.001 |
cm_r1c1 | 0.000 | 0.008 |
em_r1c1 | 0.006 | 0.001 |
mean | 20.896 | 0.050 |
betaSelf_r1c1 | 0.013 | 0.009 |
betaCoTwin_r1c1 | 0.008 | 0.009 |
umxGxE
currently assumes twins share the moderator (e.gall(T1mod == T2mod)==TRUE
Under the assumption that the latent causes of the moderator do not exert a moderated influence on the DV, it is legitimate to allow non-shared moderators, as long as both twins have their co-twin's moderator regressed out of their DV score (in addition to regressing out their own score on the moderator from their DV).
i.e. the moderated univariate model is run on
residuals(DV ~ T1mod + T2mod)
TODO
notes
umxGxE
used to support regressingmod
andmod^2
out of the DV. Now it will support onlymod
, notmod^2
when in non-shared moderator mode. This is not a great loss as most people don't regress powers of the moderator, and can be added if needed, perhaps as part of a highly flexible formula-based moderator approach, like #116